2.1 Theory of multi component hot wire measurements
2.1.3 Coordinate transformation
Toreducethenumberofunknownsinequation10itisnecessarytoexpress
S i
andU bi
as functions of U,V and W which are dened in the probe xed coordinatesystem,(
x p , y p , z p
). Figure3denesthecoordinatesystemandtheanglesneededto relatetheprobewirestothecoordinatesystem.
φ i
istheangle betweentheprojectionof wireiin the(y p − z p
)planeandthey p
axis. Figure4showstheprojectionofthewires inthe (y p − z p
)planeandthecorresponding
φ
angles.If the wires are placed in a perfect triangle, the values of the angles will be
90 ◦ , 330 ◦
and
210 ◦
respectively. Twovelocitycomponentsaredenedinthe(
y p −z p
)plane,
U bi
is the binormal cooling of wire i andU T P i
is the projection of thetangentialcoolingvelocityofwirei,tpreferstotangentialprojection.
U T P
andU b
canbecalculatedfortheindividualwires. TheyarefunctionsofV,Wand
φ i
.U T P i = V cos φ i + W sin φ i
(13)U bi = V sin φ i − W cos φ i
(14)Thevelocitycomponentinthenormal-tangentialplaneofwirei,
S i
,isafunctionofU andtheprojectionofthetangentialcoolingvelocity.
S i 2 = U 2 + U T P i 2
(15)Substitutingfor
U T P
inequation15yieldsS asafunction ofU,VandW.Figure4: Velocitiesandanglesin the(
y p − z p
)projectionS i 2 = U 2 + ( V cos φ i + W sin φ i ) 2
(16)Theow angle
α
in thenormal tangentialplane must also bedened. FromgureXanexpressionfor
α
iseasilyfound.Figure 5: Denitionof
α
α = arctan U T P
U
= arctan
V cos φ i + W sin φ i U
(17)
Equations13,15and17maybesubstitutedintoequation10toyieldthenal
equationforthreedimensional owoverwirei.
U i 2 cos 2 ( α ei ) = ( U 2 + ( W cos( φ i ) − V sin( φ i )) 2 )
(18)+cos 2
α ei + arctan
W cos φ i − V sin( φ i ) U
+( W sin( φ i ) + V cos( φ i )) 2
Forthe three wireprobea set of three equations with three unknownsis
ob-tained.
Forthe individual wires the eective angle
α e
must be found. This obtainedbyplacingtheprobeinauniformowwithavelocityS, andmeasuringtheresponse
formultipleyawangles,
α
,whilethepitchangleβ
isheldconstantatzero.Equation8describestherelationbetweentheowvelocityandthevelocitymeasuredbythe
wire,U(E).TheratiobetweenSandU(E)canthenbefoundfromequation8.
U ( E )
S = cos( α e + α )
U ( E ) cos( α e )
(19)Theresultsfrom measurementsat dierentangles
α
canthenbecurvttedtoequation19byadjusting
α e
toobtainthebestttothedatapoints. Atypicalset of calibrationanglesisα = − 20 : 5 : 20
. Insection 2.1.2α e
waspresentedasthegeometric angle betweenthe wirenormalin thenormal-tangentialplaneand the
x p
axis. This is notentirelytrue,α e
will also beafunction of the properities of the individual wire and most importantly of the ow angle. The eective angleapproachassumesthat theeectiveangleisconstant,thisishowevernottruefor
large ow angles. A litterature review by Lekakis [5] found several estimates of
the limits for x-wireprobesranging from
± 12 ◦ − ± 20 ◦
. Russ and Simon found
that the range of valid angles were larger for three-wire probes, in the range of
± 30 ◦
(reportedinthelitteraturereviewofAanesland[1]).
2.2 Probe volume and frequency response
The spatialresolution isan importantproperty ofameasurementtechnique. All
measurementtechniqueshavealowerlimitforspatial resolution, the variationis
large. Thespatialresolutionofapitotequalsthediameteroftheprobeatleast,for
alaserdopplerthesizeofthecrossectionofthelaserbeamsisthelimit,inparticle
imagevelocimetryitwilldependonthewindowsizeandoverlappingamongother
factors.
ForasinglesubminiaturehotwireLigraniandBradshaw[6]foundtheideal
ra-tiobetweenwirelengthanddiametertobeapproximately
L/D > 260
forL < 1 mm
for measurentsin aturbulent boundarylayer. Forlonger wires 'eddy averaging'
wasreported. Thewires usedin thisprojectis notclosetothedimensionsofthe
wiresusedbyLigraniandBradshaw,andcanthereforenotbeexpectedtoresolve
thesmallestscalesintheowaccurately. Thephysicalsizeofthethreewireprobe
willhoweverbeagreaterlimitingfactorthanthedimesionsoftheindividualwires.
Avelocitygradientacrossthemeasurementvolumeoftheprobewillmeanthat
thewiresintheprobeexperiencedierentvelocities. Inaowwithalargevelocity
gradient, i.e. close to awall, this can resultin large dierences acrosstheprobe
volume and distort the result. The size of theprobevolume will therefore limit
howlargegradientswhichcanbemeasured.
Theresponseoftheindividualwires isalso importantto obtainagoodresult.
Ifthefrequencyresponseofthehotwireanemomtersaredierent,someturbulent
componentscanbeoverestimated. If forexamplethegoalof anexperimentis to
validate whether aowis isotropicornot,adierencein frequency response can
responses.Toreducetheeectofthis,thesignalsshouldallbelteredatthesame
cutofrequency.
2.3 Turbulent pipe ow
A conned ow such as apipe ow will develop until a steady state solution is
reached. Assumingthattheowenteringthepipeisuniform, theboundarylayer
willimmediatelystarttogrowatthewall. Thenalsteadystatesolutionisreached
whentheinviscidcoreisgone,theowisthensaidtobefullydeveloped. Theform
ofthevelocityprolewilldependonwhethertheowisturbulentorlaminar,the
wallroughnessand thepressuregradient.
Fully developedturbulentpipeowwill exhibitcertaincharacteristics. Inthis
sectionabriefreviewofsomeofthese characteristicsisgiven.
2.3.1 The pressuregradient
Auniformowenteringapipewillberetardedbytheshearstressfromthewalls.
The pressure gradient will be greatest in the beginning, and gradually decrease
untiltheowis fullydeveloped. Atsteadystatethedriving forceof thepressure
gradientwillbalancetheshearstressonthewall.
∂P
∂x = τ w 4
D
(20)Thewallshear stress canberelated to the wall-friction velocity,
u ∗
, which isanimportantparameterin pipeow.
τ w = ρu ∗ 2
(21)Bycombiningequation20and 21thewall-frictionvelocitycanbe foundfrom
thepressuregradient.
u ∗ 2 = ∂P
∂X D
4 ρ
(22)2.3.2 Mean velocityprole
Aturbulentpipeowwill consistofthreeregions.
•
Aninnerlayerclosetothewallwhere viscousshearisdominating•
Anouterlayerwhere turbulentshearisdominating•
Anoverlaplayermergingthetwolayerstogether,wherebothtypesofshearisimportant.
commonapproachisto identifytheimportantparametersinthedierentregions
and apply dimensional analysis. In the inner region the velocity is assumed to
depend on the wall shear, uid properties and the distance from the wall. Free
stream conditionsareassumed notto be important. The wallshear will however
dependonfreestreampropertiessuchasthepressuregradient.
u ¯ = f ( τ w , ρ, µ, y )
(23) Dimensionalanalysisyieldstwodimensionlessparameters.u ¯
Thetwodimensionlessgroupsaredenoted
u +
andy +
respectively,givingu + = f ( y + )
. In the inner viscous shear dominated region turbulent shear can be ne-glected. Analysisof themomentum equation will thenyield thatu + = ( y + )
, see e.g. White[12]. Intheouterregionofthepipeowthevelocitynolongerdependson viscous shear, but on the freestream pressure gradient and the radius of the
pipe,R.
U cl − u ¯ = f ( τ w , ρ, R, ∂P
∂x ) , y
(25)Dimensionalanalysisyieldsthreedimensionless groups.
U cl − u ¯
Somewherebetweentheinnerandouterlayer,thetwolayersmustmerge,giving
thesamevelocity. Atagivenaxialpositioninthepipe,theshapeofgisassumed
tobeafunctionof
ξ = R
τ w
∂P
∂x
. Theoverlaplawforagivenξ
canthenbefoundbymanipulating equation24and 26.
u ¯
Thetwo regionscanonly bemerged ifthe fand g are logarithmic functions.
Theresultingrelationcan bewritten bothintermsofinnerandoutervariabels.
u ¯
Dierent valueshavebeen suggestedfor theconstants
κ
and B,but theyareconsideredto benearlyuniversial.A willdepend on
ξ
.In a fully developed pipe ow the only mean velocity component is that in the
streamwisedirection,U.Insection2.3.2itwasshowedhowUvariesasafunction
ofy. Theshearstressesin aowarecloselylinkedtothemeanvelocitygradients.
ThegeneralizedBoussinesqeddyviscosityhypothesissuggestsarelation.
u i u j = ν T ∂U i
∂x j − 2
3 ρkδ ij
(30)Basedonequation30,onecanmakesomassumptionsonthemagnitudeofthe
shear stressesin a pipe ow. Theonly mean velocity gradient is
∂U
∂y
, one wouldthereforeexpect
uv
tobethedominantshearstressintheow.The variation of
uv
as a function of y can be found from manipulation of the Reynolds averagedNavier-Stokesequations. Equations 31 and 32 show thesimplied RANSequationsforthepipeow.
∂p
Byintegratingequation32withrespecttoy,from0toyanexpressionforthe
pressureatagivenycoordinateisfound.
P
ρ + ¯ v 2 = P 0
ρ
(33)Ifonetakesthederivativeof thepressurewith respectto x onewill nd that
∂P
∂x
isconstantwithrespecty,seeingthatv 2
isnotafunction ofx.∂P
∂x = ∂P 0
∂x
(34)The equation in the x-direction can be integrated in the same manner, with
respect to y from 0 to y. By using the fact that
dP
dx
is constant the followingexpressionisfound. theknownsituation at thecenter linean expressionforthe variationofthetotal
stresscanbefoundasafunctionofy.
−uv + µ
Equation37providesvaluableinformationabouthow
uv
vary asafunction ofy. In the inviscid region viscous shear stress is neglible and the turbulent shear
stress is expected to vary linearly with respect to y. And at the center line all
shearstressesareexpectedtobezero.
2.3.4 Turbulentnormal stresses
Boussinesq estimatesthe normalstressesto beonethird ofthe turbulentkinetic
energyk. Inturbulentpipeowthatisnotthecase. Equation30assumesisotropic
and homogeneous turbulence, but in a shear ow the production of the normal
stresses will vary. In the case of a turbulent ow the turbulent kinetic energy
equationin theaxial directionwillbetheonlyonewithaproductionterm.
P roduction = −uv ∂U
∂y
(38)Theproductiondependsonthemeanowgradient,asmeanvelocityin they
andzdirectioniszeroforafullydevelopedpipeowtheproductionoftheturbulent
normalstressesis zero. Thisdoesnotmeanthattheothernormalstresseswillbe
zero. Energy is transfered from
¯ u 2
tov ¯ 2
andw ¯ 2
by nonlinear pressure-velocity interactions[8].2.4 Cylinder wake
ThecylinderwakeisacomplexandReynoldsnumberdependentow. Forverylow
Reynoldsnumbers,Re<49,alaminar,symmetricalandsteadyrecirculationregion
is presentbehindthe cylinder. As the Reynoldsnumberincreaseslaminar vortex
shedding will begin. When the Reynoldsnumber reaches about 194 streamwise
vorticesbegintoform[4]. Uptoabout
Re D = 1000
theStrouhalnumberincreases [11]. TheStrouhal numberis dened astheratio betweenf D
and U, where fisthevortexsheddingfrequencybehindthecylinder.
St = f D
U
(39)For
Re D > 1000
the Strouhalnumber startto decrease untill it stabilizes for10000 < Re D < 100000
atavaluecloseto0.21[11]. TheregionfromRe D = 1000
to
Re D < 200000
isnamedthesubrcriticalrange[13]. Inthesubcriticalrangethe boundarylayeronthecylinderremainslaminar. IftheReynoldsnumberincreasesfurthertheboundarylayerstartstodevelopfromlaminartoturbulent,movingthe
pointoftransitionupstreamandtheseparationpointdownstream. Thisresultsin
reduceddragandanarrowedwake.
Unlike apipe ow, the wake is continually evolving. The mean velocity eld
will continue to developuntil free stream conditions are reached. Momentum is
continuallytransportedtowardsthecenterofthewakewherethevelocitydecitis
form yieldsthefollowing.
∂V
∂y = − ∂U
∂x
(40)The continuity equation tells us how the gradient of V with respect to y is
expected to vary. Far from the centerline
dU
dx
will benegative, sincethe wakeisexpanding. Inthisregion
dV
dy
willbepositive. Closertothecenterofthewakeweexpect
dU
dx
tobepositive,dV
dy
mustthereforebenegative.By performing an order of magnitude analysis, the x-direction Reynolds
av-eragedNavier-Stokesequation canbe simplied considerably. Thetwodominant
terms arethe U gradientwith respectto x and crossectionalgradientof the
tur-bulentshear stress
uv
.U ∂U
∂x = − ∂
∂y ( uv )
(41)Fardownstream from thecylinder
( x/D > 80)
theowcanbeassumedto be self-preserving[ref], which means that theshape ofthe proleis preservedalongthex-axis. Theshapeoftheprolecanbefoundbystartingwithequation41and
makingsomeadditionalassumptions.
3.1 The hot-wire probe
The probe consist of three wires on six supporting prongs, asshownin gure3.
Thechosengeometryisdenedbytwoproperties:
•
Theprojectionofthewires inthey p − z p
planeisatrianglewith60degreeangles
•
Thewires areinclinedanangleα e = 35 . 26 ◦
relativetothe
y p − z p
planeThesepropertiesgiveageometrywherethewires areorientatedperpendicular
to one another. The geometry is the same as recommended by Aanesland [1].
It was chosen to reduce the probe volume and give a good cooling response in
all directions. The probesare manufacturedto t theabovedescription,but the
angleswillneverbeexactlycorrect. Theeectiveanglesmustbefoundtroughthe
procedure described in section 2.1.4. By taking a picture of the probe trough a
microscope,theorientationofthewiresin the
y p − z p
planecanbefound, such aphoto can be seenin gure 6. Theangle
φ 1
is used to relate therotationof theprobein the
y p − z p
planetocoordinatesystem.Figure6: Pictureofthe
y p − z p
planetakentroughamicroscopeThelengthofthesupportingprongsischosensuchthattheowoverthewires
are notinuencedbytherest of theprobes. It isalso important that theprongs
are nottoolong,asthiscancausevibrationswhich inturn canbe interpretedas
aturbulentvelocitycomponent.
Aplatinum(90
%
),rhodium(10%
)alloyisusedinthewire. Thisalloygives agoodoxidationresistance,relativelyhightensilestrengthbuthasarelativlylowThe diameter also inuences the sensitivity of the probe. In this project a wire
with d =
5 µm
is used. This is arelativley thick wire, which reduces sensitivity but increases mechanicalstrength. Thelength ofthe wirebetween theprobesisapproximatly4mm,ofthis approximatly1.75mmof thecoatingonthewirehas
beenetchedawayin thecentre. Thisgives
l/d ≈ 350
. Toreduce theinterference ontheowfromthesupportingprongsthedistancebetweenthesupportingprongsshould notbetosmall,theshapeoftheprongtipswillalsoaect theow[2].
Thecrossectionofthemeasurementvolumeis
≈ 5
mm,andthespatial resolu-tionoftheprobeisthereforassumedtobe5mm.3.2 Measurement chains
Figure7describesthemeasurementchainintheexperiment.
Figure7: Measurementchain
Thehot wireanemometersare optimized for
1 µm
notfor5 µm
which is usedin this experiment. For the initial measurement setup a high frequency
distur-bance appearedonthesignalathighvelocities
( > 12 m/s )
. This isaresultofthe inability of the control circuitto regulate thewire voltage. It could to acertaindegreebehelpedbychangingthebiassetting ontheanemometer. Thisincreased
the dampingin the control loop at the cost of a lowerfrequency response. For
thepipemeasurementsthiswassucient,in thecylinder wakehoweverthelarge
uctuationsrequired that higher velocities couldbe measured. The solutionwas
to extendthecable, increasing
R cable
,and therebyincreasing thedampingintheloop.
3.3 Signal sampling rate
The sampling rate must be set accordingto the timescale of the smallesteddies
ofinterest. Whentherangeoftimescalesexpectedisunknown,thesamplingrate
mustbesetaccordingtothesmallesttimescaleoncanexpect. Kolmogorovsmicro
notbeendoneinthisproject.
Thelimitingfactorforthesamplingrateisinthiscaseisthefrequencyresponse
oftheanemometers. Thefrequencyresponsewasfoundtovarybetweenthewires
from approximatly6.3 kHzto 8.0 kHz. A lowpass ltercut o frequencyof 6.5
kHz waschosen. Thesamplingrate shouldbeset accordingto thesamplingrate
theoremorNyquistcriteria,whichstatesthatthesamplingrateshouldbegreater
thantwicethemaximumfrequencyexpectedtoavoidaliases[10].
A suitablesampling time should bechosensuch that repeated measurements
givethesameresult,averagingoverrelevanttimescalesintheow.
Inthecylinderwakemeasurementsasampling rateof 13kHzwasusedalong
with a sampling time of 20 seconds. For the pipe measurements the sampling
frequency wasset to 7 kHzand the sampling time to 10 seconds. This wasnot
intendedtobethenalmeasurements,but simplypreliminarymeasurements,the
reducednumberofdatapointsgavesignicantlyreduceddatasizeandwastherefore
chosenatthetime.
3.4 Data reduction program
The sampled signal from the velocity calibration, the eective angle calibration
and traverses, were stored in text les and imported into a Fortran script. The
script corrects the data for temperature change, ts polynomials to the velocity
calibration data, calculates the eective angles, and uses the calibrationdata to
calculatetimeseries ofvelocityvectorsfromthevoltagetimeseries.
The solution of the equations (Eqs. 18) was be found by using a zero point
nder. Initially afortranfunction called DNSQE from theSLATEC librarywas
used, this function had previouslybeenused byAanesland [1] with success. The
algorithmworkedneforaveragedvoltages,butconvergensproblemsarisedwhen
the turbulenttimeseries wereanalyzed. As analternativeMatlabs fzero function
wasused. TheMatlabfunctionisconsiderablyslowerthantheFortranroutinebut
it doesthe job. Simple constraintswere placed on the solutionto insure that a
physicallycorrectsolutionwasfound. TheFortranscriptusedfordataanalysisis
describedfurherinappendix6.
3.5 Pipe ow rig
ThepiperigconsistsofahydraulicallysmoothPVCpipe,with adiameterof 186
mm and a length of 83 diameters. Ten pressuretaps are mounted on the pipe,
making it easy to measure the pressuregradient. The pipe is tted such that a
traversecan be mounted ontop, making it possible to traverse the ow through
thecenterofthepipe. Velocitiesinthepiperigcouldbevariedfrom5to12.5m/s.
Thecoordinate systemusedin thepipehasits reference
( y = 0)
onthe centre lineofthepipe,yispositiveabovethecenterline,andnegativebelowthecentreline.The velocities in thepipe aredenoted
U x
,U r
andU theta
and arethe axial, radialandcircumferentialvelocitiesrespectively.
An openloop wind tunnel is used forthecylinder wakemeasurements. The test
sectionis45cmx45cmand110cmlong. Acylinderwithadiameterof47.5mm
is tted in the center of the test section, leaving 50 cm of distance downstream
for the ow to develop. Measurementsare takenat
x/D = 10
. Velocities in the windtunnelcouldbevariedfrom4to30m/s.Inthewindtunnelthecentreofthewakeisthereference
( y = 0)
inthe coordi-nate system,yispositiveabovethecenterline,andnegativebelowthecenterline.U,VandWaretheaxial,vertical,andtransversevelocitycomponentsrespectively.
4.1 Calibration and testing
4.1.1 Velocity calibration
Inthepiperigthevelocitycalibrationwasperformedforvelocitiesbetween5and
12.5m/s. Athird order polynomialt to thecalibrationdata includingthezero
velocitypointgavearesidualoftheorderof
10 − 1
,whileasecond orderttothevelocitypointgavearesidualoftheorderof